The Solution of the Metric STRESS and SSTRESS Problems in Multidimensional Scaling Using Newtons Method (Q1297863)

Multidimensional scaling (MS) is well known in the analysis of psychological data. The authors consider its application to the problem of molecular structure reconstruction from the information about interatomic distances. The number of objects in this problem is much greater than in typical psychological problems and the distances are measured much more accurate. The authors discuss known algorithms of MS based on the iterative gradient optimization of STRESS and SSTRESS criteria. They propose an original algorithm which uses a globalized Newton's method. They claim that it should have better convergence rates. Results of numerical experiments on comparisons of this method with standard MS algorithms are considered.